Commit History
-
Update src/model/embedding.rs
Unverified signatureCo-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>668baf3 -
Update src/text/phoneme.rs
Unverified signatureCo-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>a27c268 -
Update src/text/phoneme.rs
Unverified signatureCo-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>415ac42 -
Convert IndexTTS to pure Rust implementation
ClaudeUnverified signatureComplete rewrite of IndexTTS from Python to Rust for maximum performance: Core Features: - Audio processing with mel-spectrogram computation using RealFFT - STFT/iSTFT implementation with proper windowing - Mel filterbank creation with triangular filters - Audio I/O for WAV files using hound - Sample rate resampling with rubato Text Processing: - Text normalization (Chinese/English punctuation handling) - BPE tokenization with vocabulary support - Phoneme conversion system - Multi-language detection (Chinese, English, Mixed) Model Inference: - ONNX Runtime integration with session management - Speaker encoder for voice cloning - Emotion encoder with 8-dimensional control - Semantic encoder for audio codes - GPT model wrapper with sampling strategies Vocoder: - BigVGAN neural vocoder (fallback implementation) - Snake activation functions for neural synthesis - Griffin-Lim as simple fallback vocoder Pipeline: - Complete TTS synthesis orchestration - Configurable synthesis options - …2bbfbb7